Burstsort and its variants are cache-efficient algorithms for sorting strings. They are variants of the traditional radix sort but faster for large data sets of common strings, first published in 2003, with some optimizing versions published in later years.
Burstsort and its variants are cache-efficient algorithms for sorting strings. They are variants of the traditional radix sort but faster for large data sets of common strings, first published in 2003, with some optimizing versions published in later years.
Burstsort algorithms use a trie to store prefixes of strings, with growable arrays of pointers as end nodes containing sorted, unique, suffixes (referred to as buckets). Some variants copy the string tails into the buckets. As the buckets grow beyond a predetermined threshold, the buckets are "burst" into tries, giving the sort its name. A more recent variant uses a bucket index with smaller sub-buckets to reduce memory usage. Most implementations delegate to multikey quicksort, an extension of three-way radix quicksort, to sort the contents of the buckets. By dividing the input into buckets with common prefixes, the sorting can be done in a cache-efficient manner.
Discovered by embedding cosine similarity (sentence-transformers MiniLM, 384-dim).